Combinatorial and Computational Geometry

MSRI Publications
Volume 52, 2005

Extremal Problems Related to the

SylvesterGallai Theorem
NIRANJAN NILAKANTAN

Abstract. We discuss certain extremal problems in combinatorial geometry, including Sylvesters problem and its generalizations.

1. Introduction
Many interesting problems in combinatorial geometry have remained unsolved
or only partially solved for a long time. From time to time breakthroughs are
made. In this survey, we shall discuss the known results about some metric and
nonmetric problems. In particular, we shall discuss the SylvesterGallai problem
and the DiracMotzkin conjecture on the existence and number of ordinary lines,
the Dirac conjecture on the number of connecting lines, and the problem of
distinct and repeated distances. The main focus will be on versions of these
problems in the Euclidean and real projective plane.
The method of allowable sequences will be described as a tool to give purely
combinatorial solutions to extremal problems in combinatorial geometry.

2. Sylvesters Problem
Sylvester [1893] posed a question in the Educational Times that was to remain
unsolved for 40 years until it was raised again by Erdos [1943]. Then it was soon
solved by Gallai [1944], who gave an affine proof. More followed: Steinbergs
proof in the projective plane and others by Buck, Gr
unwald and Steenrod, all
collected in [Steinberg et al. 1944]; Kellys Euclidean proof [1948], and others,
including [Motzkin 1951; Lang 1955; Williams 1968].
We give the following definitions before we state the problem and its solutions.
Let P be a finite set of 3 or more noncollinear points in the plane. Let F be
a finite collection of simple closed curves in the real projective plane which do
479

480

NIRANJAN NILAKANTAN

not separate the plane, every two of which have exactly one point in common,
where they cross. F is known as a pseudoline arrangement.
Connecting line: a line containing two or more points of P .
Ordinary line: a connecting line which has exactly two points of P on it.
Vertex: an intersection of two or more lines of a straight line arrangement or
pseudolines of a pseudoline arrangement.
Ordinary point: a vertex which is the intersection of exactly two lines or two
pseudolines.
Sylvester asked for a proof of the statement that every set P of noncollinear
points always determines an ordinary line. In the dual, one has to show that
any straight line arrangement in which not all lines are concurrent has an ordinary point. By the principle of duality, proofs for point configurations carry
over trivially into proofs for line arrangements and vice versa. The canonical
correspondence maps the point (a, b) to the line y = ax + b.
Levi [1926] introduced the notion of a pseudoline defined above. A natural
question to ask is whether every pseudoline arrangement in which not all pseudolines are concurrent has an ordinary vertex. This is more general than the
question of whether every straight line arrangement has an ordinary vertex, since
every straight-line arrangement has an equivalent pseudoline arrangement, but
there exist unstretchable pseudoline arrangements [Gr
unbaum 1970; Goodman
and Pollack 1980b].
Solutions to Sylvesters problem. We now show some of the techniques used
to solve Sylvesters problem in the both the primal and dual versions, and in the
Euclidean as well as the projective plane.
Gallais proof (affine). Choose any point p1 P . If p1 lies on an ordinary line,
we are done, so we may assume that p1 does not lie on any ordinary line. Project
p1 to infinity and consider the set of lines containing p1 . These lines are parallel,
and there are at least two such lines. Let s be a connecting line not through p1
which forms the smallest angle with the parallel lines:
s
p1

We assert that s is ordinary. If not, it must have at least 3 points p2 , p3 , p4 , as

in the figure at the top of the next page. The connecting line through p1 and p3
has another point p5 , since it is not ordinary (this point is shown in two possible
positions in the figure). Then, either p5 p2 or p5 p4 forms a smaller angle with
the parallel lines than s, contradicting the hypothesis that s forms the smallest
angle.

PROBLEMS RELATED TO THE SYLVESTERGALLAI THEOREM

481

s
p2
p3

p5

p5
p1

p4

Kellys proof (Euclidean). We have the set P of points not all collinear and the
set S of connecting lines determined by P . Any point in P and any connecting
line not through the point determine a perpendicular distance from the point to
the line. The collection of all these distances is finite, because P and S are finite,
so there is a smallest such distance. Let p P and s S be a nonincident
pair realizing this smallest distance, and let q be the foot of the perpendicular
line from p to s :
q

p1

p2

Then s is ordinary; otherwise it would contain three points of P , at least two

of them lying on the same side of q. Let these two points be p1 and p2 , with p1
between q and p2 . Now the distance from p1 to the connecting line p p2 would
be less than the distance from p to s , giving a contradiction.
Steinbergs proof (projective). With S and P as above, take any p in P . If p lies
on an ordinary line we are done, so we may assume that p lies on no ordinary
line. Let l be a line through p that is not a connecting line, that is, one that
contains no point of P apart from p. Let Q be the set of intersections of l with
lines in S, and take q Q next to p (meaning that one of the open segments
determined by p and q on the projective line l contains no element of Q). Let
s be a line of S through q; then s must be ordinary. Otherwise, there would be
three points of P on s, say p1 , p2 , p3 (arranged in that order in s \ {q}; note that
q is not in P , by our choice of l):
p
p4

p1

p2

p3

p4

The line through p and p2 would then contain another point of P , say p4 , since
p lies on no ordinary line; then p1 p4 or p3 p4 would meet the forbidden segment
pq (see the figure where two possibilities for p4 are shown).

482

NIRANJAN NILAKANTAN

The DiracMotzkin conjecture. Having determined the existence of an

ordinary line (or point, in the dual problem), attention was turned to the problem
of establishing the number of ordinary lines (or points). For P an allowable set
of points one not all of whose elements are collinear let m(P ) denote the
number of ordinary lines determined by P . Define
m(n) = min m(P ),
|P |=n

where P ranges over all allowable sets of points of cardinality |P | = n.

De Bruijn and Erd
os [1948] proved that m(n) 3, and this was proved again
by Dirac [1951], who conjectured that there were at least bn/2c ordinary lines. In
a different context, Melchior [1941]
proved again the m(n) 3 bound. Motzkin
[1951] improved this to m(n) > 2n 2. Kelly and Moser [1958] improved the
lower bound to 3n/7. Kelly and Rottenberg [1972] proved the same result for
pseudoline arrangements. In 1980, Hansen gave a lengthy proof of Diracs
bn/2c conjecture, but it was found to be incorrect by Csima and Sawyer [1993],
who nonetheless proved that there exist at least 6n/13 ordinary lines.
Creating point configurations with few ordinary lines is hard. When n is odd,
we know of configurations where the conjecture is tight only when n = 7 and
n = 13. The former is shown by the KellyMoser configuration [1958]:

and the latter by the CroweMcKee configuration [1968]. The Boroczky configurations [Crowe and McKee 1968] are valid for all even n; they are most easily
visualized dually here is the case n = 12, with dots marking ordinary vertices:

PROBLEMS RELATED TO THE SYLVESTERGALLAI THEOREM

483

Solutions to the generalized problem. We now outline the techniques used

in the progress towards settling the conjecture.
pencil: a collection of lines all of which intersect at a single point.
near-pencil: a collection of lines all but one of which intersect at a single point.
According to the moments convenience, we assume given either some arrangement L of lines not forming a pencil or near-pencil, or a configuration of points
not all collinear. We seek to prove a lower bound for the number of ordinary
points in the first case, and ordinary lines in the second.
Melchiors proof of the existence of 3 ordinary points. The lines of L partition
the real projective plane into polygonal regions. Let V, E and F denote the
number of vertices, edges and faces in the partition. By Eulers formula,
V E + F = 1.
Let fi denote the number of faces with exactly i sides and vi the number of
vertices incident with exactly i lines. Since the lines are not all concurrent,
every face has at least three sides, so f2 = 0. Then,
X
X
X
X
ivi .
ifi = 2
fi , 2E =
vi , F =
V =
i2

i3

i3

i2

This implies that

3 = 3V E + 3F 2E = 3
=

X
i2

(3 i)vi +

X
i3

X
i2

vi

ivi + 3

i2

X
i3

fi

ifi

i3

(3 i)fi ,

and hence that

v2 = 3 +

X
i4

(i 3)vi +

X
i4

(i 3)fi 3 +

X
(i 3)vi .
i4

Thus, any finite set of nonconcurrent lines has at least 3 ordinary points.

Motzkins proof of the existence of O( n) ordinary lines. Consider a point

p P not lying on any ordinary line. (If there is no such point, there are at
least n/2 ordinary lines and we are done.) Consider the set of connecting lines
not passing through p. These partition the plane into regions, and p lies in one
of these, which is called its cell C. If p has at least 3 lines on the boundary of
its cell, then all the lines in the boundary of the region containing p must be
ordinary.
It is easy to see that no point of P can lie on the edges of the cell C. Suppose
one of the lines l on the boundary of the cell is not ordinary, that is, l has 3
points p1 , p2 , p3 labeled so that p1 , x separate p2 , p3 , where x is a point on l not
in P on the boundary of C (see figure on the next page). The line pp1 is not
ordinary by hypothesis, and therefore contains a point q of P . But then either

484

NIRANJAN NILAKANTAN

q
p
x
p3

p2

p1

qp2 or qp3 cuts the cell C, contradicting the fact that C is the polygonal region
containing p.
Thus, the ordinary lines partition the plane into polygonal regions, and all
the points which do not lie an any ordinary line lie in one of these regions. It is
easy to see that no region can have more than one

point.
Now, m ordinary lines determine at most m
2 + 1 regions, and can have at
most 2m points
of P on them. Since every point is
on an ordinary line or in a
+
1
+
2m

n,
implying
that
m

cell, we have m
2n 2.
2
Kelly and Mosers proof of the existence of 3n/7 ordinary lines. Let P be the set
of points and S the set of connecting lines. We denote a generic point by p and a
generic line by s. The set of lines of S which do not go through p subdivide the
plane into polygonal regions. p is contained in one of these polygonal regions,
which is called its residence.
Neighbor of p: a line of S containing the edges of the residence of p.
Order of p: the number of ordinary lines passing through p.
Rank of p: the number of neighbors of p which are ordinary lines.
Index of p: the sum of its order and rank.
Theorem 1. If a point q has precisely one neighbor , then S is a near-pencil .
This is because the neighboring line is the only line which does not pass through
q, and all the other lines pass through q:

Theorem 2. If a point p has precisely two neighbors, then S is a near-pencil .

The lines of S that do not pass through p form a pencil, or else p would have at
least three neighbors. Let q be the vertex of the pencil. Let si and sj be two

PROBLEMS RELATED TO THE SYLVESTERGALLAI THEOREM

485

lines through q and pi and pj be points on si and sj respectively, different from

q. The connecting line through pi and pj does not pass through q and therefore
passes through p. Thus, only one line of S passes through p, and all the rest
pass through q, as in the previous figure.
As a consequence of the previous two theorems, we have:
Theorem 3. If S is not a near-pencil , each point of P has at least three neighbors.
Theorem 4. If the order of p is zero, every neighbor of p is an ordinary line.
This was proved in [Motzkin 1951]; we gave the proof on page 483.
Theorem 5. Any point of P not of order two has index at least three.
If the order is zero and S is not a near-pencil, the rank is at least three. If the
order is at least three, there is nothing more to prove. If the order is one, the
rank is at least two and the correct proof of this was given by Dirac in his review
of Kelly and Mosers article [Dirac 1959].
Theorem 6. If a line s of S is a neighbor of three points p1 , p2 , p3 , then the
points of P which lie on s are on the connecting lines determined by p1 , p2 , p3 .
Three points that have a common neighbor cannot be collinear: if p1 , p2 separate x, p3 , where x is the intersection of s with p1 p2 , then s cannot lie on the
boundary of p3 s cell. Let the intersections of p1 p2 , p2 p3 , p3 p1 with s be x3 , x1 ,
x2 respectively. If p is a point of P on s such that xi xj separate xk p, then ppi
and ppj separate s from pk . Here, i, j, k is some permutation of 1, 2, 3.
p1
p2
x3

x1

x2
p3

This implies the following.

Theorem 7. A line l of S is a neighbor of at most four points.
Suppose l was the neighbor of five points p1 , . . . , p5 . Looking at p1 , p2 , p3 , we
see that at least 2 of x1 , x2 , x3 must be elements of P . Assume that x2 , x3 are
elements of P . However, neither x2 nor x3 can be on the lines p1 p4 or p1 p5 . This
means that one of the points of P on l is not on the connecting lines of the set
p1 , p4 , p5 , implying that l is not a neighbor of one of the three points.
Theorem 8. If Ii is the index of the point pi , then
n

1X
Ii
6 i=1

486

NIRANJAN NILAKANTAN

Since each ordinary line can be counted at most six times four times as a
neighbor and twice as being incident with each of its points the sum of the
index over all the points is greater than six times the number of ordinary lines.
Theorem 9. m 3n/7.
Let k be the number of points of order 2. Then
m

3n k
3(n k) + 2k
=
,
6
6

which leads to 6m 3n k 3n m since m k (trivially). Hence m 3n/7.

Proof by Csima and Sawyer. Csima and Sawyer improved upon Kelly and Moser
by showing that except for pencils and the KellyMoser configuration the number
6
n, with eqaulity
s of ordinary points in a configuration of n lines is at least 13
occuring for the McKee configuration. They generalize the KellyMoser proof
in the following way. In the KellyMoser proof, the sum of the indices of each
point was compared to the six times the number of ordinary lines to get the
desired bound. In the CsimaSawyer result, the index is a weighted sum of the
order and the rank. The following is a sketch of their proof for an arrangement
of lines, and works for arrangements of pseudolines as well.
Attached: An ordinary point which not on a line but associated to it, by
proximity. For instance, in the proof of Kelly and Moser, the ordinary lines
on the boundary of the cell of a point are attached to it.
Type of a line l: The pair T (l) = (, ), if there are exactly ordinary points
on l and ordinary points attached to l.
-weight of a line l of type (, ): : the number w (l) = + .
Theorem 10. Suppose is a finite configuration of lines in the real projective
plane having two lines of type (2, 0) that intersect in an ordinary point. Then
is the KellyMoser configuration.
Theorem 11. Apart from pencils, if T (l) 6= (2, 0), then w1 (l) 3.
This is a restatement of a theorem of Kelly and Moser, which asserts that the
index of a point which is not of order two is at least three.
Theorem 12. If l1 and l2 have an ordinary intersection in any configuration
other than pencils, then w1 (l1 ) + w1 (l2 ) 5.
Theorem 13. Except for pencils and the KellyMoser configuration, s
Partition the ordinary points into the sets
= ordinary points that lie on a line of type (2, 0),
= ordinary points that do not lie on a line of type (2, 0).

6
13 n.

PROBLEMS RELATED TO THE SYLVESTERGALLAI THEOREM

487

and the lines into sets of bad, good and fair lines:
B = lines l of type (2, 0),
G = lines l that contain a point in but l
/ B,

F = lines l that do not contain a point in .

The set G is further partitioned into sets

Gj = lines l in G which contain exactly j points of

Consider two lines l and m. If their intersection is in , we can assume without
loss of generality that l B. Then m has a 1-weight of at least three, and lies in
G. Thus, each point in appears on exactly one line from B and one line from
G. If B = |B|, G = |G|, F = |F|, and Gj = |Gj |, we have
X
X
G=
Gj
Gj = || = 2B.
j

j1

If l G1 , then T (l) = (, ) (1, 0), and w1 (l) = + 3, and since 1,

we have w (l) = + + 2. If l G2 , then w (l) 2 + 1. If l Gj for
j 3, then w (l) j. If l B, then w (l) = 2, and if l F, then w (l) 3.
Thus,
X
X X X
X
w (l)
+
w (l) +
w (l) =
l

lB

mGj

lF

2B + ( + 2)G1 + (2 + 1)G2 +

jGj + 3F

j3

X
jGj ) + 2G1 + G2 + 3F
= 2B + (
j1

= (4 2)B + 3G1 + 3G2 +

jGj + 3F

j3

(4 2)B + 3G + 3F
Choosing =

5
4

we get,
X
w5/4 (l) 3B + 3G + 3F = 3n.
l

Consider a matrix with rows labeled by the lines l and columns labeled by
the ordinary points. If the ith line is incident with the j-th ordinary point, the
(i, j)-th entry of the matrix is 45 . If the j-th point is attached to the i-th line,
the (i, j)-th entry is 1. All other entries are zero.
An ordinary point P is attached to at most four lines. Therefore, the column
P
sum is at most 2( 45 )+4 = 13
l w5/4 (l)
2 . The sum over all the rows is exactly
3n. Consequently,
X
w5/4 (l) 13
3n
2 s.
l

488

NIRANJAN NILAKANTAN

3. Allowable Sequences
The notion of allowable sequences has proved very effective in determining
the combinatorial classification of configurations of the plane.
A configuration of n points is an ordered n-tuple of distinct points in the
plane. The points are labeled 1, 2, . . . , n. Given a configuration C and a directed
line l which is not orthogonal to any line determined by two points of C, the
orthogonal projection of C on l determines a permutation of 1, 2, . . . , n. As the
line l rotates in a counterclockwise direction about a fixed point, we obtain a
periodic sequence of permutations which is called the circular sequence of the
configuration.
Allowable sequences are circular sequences constrained by the following properties:
1. Succesive permutations differ only by having the order of two or more adjacent
numbers switched.
2. If a move results in the reversal of a pair ij then every other pair is reversed
subsequently before i and j switch again.
Allowable sequences and the Sylvester problem. The point configurations
encountered in the Sylvester problem must take into account highly degenerate
cases. Since many points may be collinear, the corresponding circular sequence
will have switches in which more than two adjacent numbers are reversed. The
problem of showing the existence of an ordinary line is equivalent to the problem
of determining whether a simple switch occurs.
History of the use of allowable sequences. Though the concept was introduced by Goodman and Pollack [1980a] to study the Erdos-Szekeres conjecture,
it has been very useful in solving a range of problems which depend mainly on
the order types of the point configuration. In particular, it has been used to
show that
every pseudoline arrangement of less than nine lines is stretchable [Goodman
and Pollack 1980b];
the number of directions determined by 2n points is at least 2n [Ungar 1982];
the number of k-sets among a set of n points is O(nk 1/2 ) [Edelsbrunner and
Welzl 1985];
the maximum number of at most k-sets is O(nk) [Welzl 1986];
pseudoline arrangements are semispace equivalent if and only if they have
the same allowable sequence modulo local equivalence [Goodman and Pollack
1984].
Properties determined by allowable sequences.
i1 , i2 , . . . , ik are collinear if and only if they switch simultaneously
i is in the convex hull of i1 , i2 , . . . , ik if and only if in every permutation in
the sequence, i is preceded by one of i1 , i2 , . . . , ik .

PROBLEMS RELATED TO THE SYLVESTERGALLAI THEOREM

489

i is an extreme point if and only if some permutation begins with i

ij is parallel to kl if and only if they both switch simultaneously
ijk turn counterclockwise if and only if ij precedes ik, written as ij ik.
ij separates k from m if and only if when ij switches, k and m are on opposite
sides of the substring ij in the permutation.

Using allowable sequences, Edelsbrunner and Welzl [1985] were able to derive
improved
upper bounds for the kset problem viz. that the number of ksets
is O(n k). Welzl [1986] generalized this result to bound the number of at
most ksets in a configuration of n points. Ungar [1982] was able to settle the
conjecture regarding the number of directions determined by a configuration of
points.
As an example of the power of allowable sequences, we give the following proof
by Ungar.
Ungars proof for the number of directions determined by 2n points.
We pay special attention to switches which straddle the midpoint of a permutation. A switch in which some indices cross the midpoint is called a crossing
move. The ith crossing move causes an increasing string straddling the midpoint
to be reversed. If di denotes the distance from the midpoint to the nearest end of
the string, then, at the ith crossing move, exactly 2di indices cross the midpoint.
Since every index must cross the midpoint, if there are t crossing moves in all,
then
2d1 + 2d2 + + 2dt 2n

since some indices can cross more than once.

Between two crossing moves, there must be at least di + di+1 1 noncrossing
moves, since we must first tear down a decreasing string of length di and build
an increasing string of length di+1 , and a decreasing string can be shortened by
at most one in a switch (an increasing string can be increased by at most one in
a switch).
Thus, the total number of switches between the first crossing move and when
this same crossing move occurs in reverse corresponds to a half period and has
P
P
(di + di+1 1 + 1) = (2di ) 2n.
This is a tight lower bound, since the regular 2n-gon determines exactly 2n
directions, as in the B
or
oczky configuration of page 482.
When the number of points is odd, say 2n + 1, the number of directions can
be shown to be at least 2n, since all but one point must cross the position n + 1
in the permutation.

4. Colored Extensions of Sylvesters Problem

Let {Pi } be a collection of sets of points, and let all points in the same set
be assigned a color. A line is monochromatic if it passes through at least two
points of the same color and no points of any other color. The following problem

490

NIRANJAN NILAKANTAN

is attributed to Graham and Newman: Given a finite set of points in the plane
colored either red or blue, and not all collinear, must there exist a monochromatic
line? Motzkin [1967] solved the problem in the dual, showing there must exist a
monochromatic point in an arrangement of colored lines. The proof is sketched
in Section 4 (page 490). Chakerian [1970] and Stein gave additional proofs.
monochromatic point: an intersection point in an arrangement of colored
lines where all the lines intersecting at that point have the same color.
Consider the following question: Does there exist for every k a set of points in
the plane so that if one colors the points by two colors in an arbitrary way, there
should always be at least one line which contains at least k points, all of whose
points have the same color? This is known to be true for k = 3, but nothing is
known for larger values of k.
Various generalizations of this problem to higher dimensions have been proposed and solved [Chakerian 1970; Borwein 1982; Borwein and Edelstein 1983;
Tingley 1975; Baston and Bostock 1978].
Clearly, we cannot insist that the monochromatic line be ordinary without
additional restrictions. In the search for ordinary lines in the colored setting,
Fukuda [1996] raised the following question. Let R be a set of red points and
B be a set of blue points in the plane, not all on the same line. If R and B
are separated by a line and their sizes differ by at most one, then there exists
an ordinary bichromatic line, that is, a line with exactly one red point and one
blue point. This conjecture is shown not to be true for small n in [Finschi and
Fukuda 2003].
Pach and Pinchasi [2000] have shown that there exist bichromatic lines with
few points.
Motzkins solution of the existence of a bichromatic point.
Theorem 14. Let S and T be two sets of nonconcurrent lines in the real projective plane colored red and blue respectively. At least one of the intersection
points in S T is monochromatic.
Suppose S and T do not define any monochromatic vertex. Then, every intersection point w of two red lines has a blue line passing through it. These lines
can be ordered so that the blue line lies in between the red lines. Since not all
the blue lines are concurrent, there is some other blue line that does not pass
through this intersection point. The new blue line forms a triangle wxy with the
two red lines, as shown here (blue = gray):
w
z
x

PROBLEMS RELATED TO THE SYLVESTERGALLAI THEOREM

491

Consider such a triangle that is minimal in the sense that it does not completely
contain another such triangle. This must exist because there only finitely many
triangles in the arrangement.
The intersection point v of the two blue lines must be monochromatic. If not,
there must exist a red line through v producing a triangle vzx of the original
type (two red lines and a blue line) which is contained in the minimal triangle,
contradicting the assumption that wxy is minimal.

5. Connecting Lines and Diracs Conjecture

Another interesting problem concerns the connecting lines of a set of P points.
Define an iline to be a connecting line containing exactly i points of P and
let ti (P ) denote the number of ilines determined by P . Also, let t(P ) =
P
i2 ti (P ). Let r(n) be the minimum over all configurations of n points of the
maximum number of connecting lines from a single point. i.e.
r(n) = minn max t(p)
P R

pP

Dirac [1951] asked whether one of the n points must always be incident with
at least b n2 c of the connecting lines. He showed that this was the best possible
by placing all the points evenly on two intersecting lines. He also proved a trivial

unbaum 1972] a list of exceptions to this formulation

lower bound of n. In [Gr
is enumerated.
Erdos relaxed the problem by asking whether it could be shown that r(n)
cn. The more general question he raised was the following. Is it true that there
exists an absolute constant c independent of k and n such that if 0 k 2 and
ti (P ) = 0 for i > n k then
ckn < t(P ) < 1 + kn
The upper bound is trivial, and the lower bound was shown by Beck [1983]
and Szemeredi and Trotter [1983], but with very small constants. Clarkson et
al. [1990] improved the constant significantly.
The question of whether t(P ) n was raised by Erdos [1943] and proved
by various people including Erdos and Hanani [Hanani 1951]. Kelly and Moser
[1958] were able to prove that
t(P ) kn 12 (3k + 2)(k 1)
if k is small compared to n and any connecting line contains at most nk points.

6. A Solution for Sylvesters Problem Using Allowable

Sequences
We now look at a simple application of allowable sequences to solve Sylvesters
problem.

492

NIRANJAN NILAKANTAN

Consider an allowable sequence of permutations of 1, . . . , n. In the first halfperiod, each permutation is obtained from the previous one by the switch of a
substring that is monotonically increasing. We shall pay special attention to the
switches involving 1 or n. We claim that the first switch involving a substring to
the right of n or a substring to the left on 1 in the permutation is simple, thus
proving the theorem.
Assume that n makes a switch before 1 makes a switch. Similar arguments
hold for 1 if this not the case. This assumption implies that the first switch
involving n does not involve 1.
Every substring switch involving n has n at the end of the substring before
switch. After the switch, the right of n in the permutation consists of a concatenation of substrings, each of which is monotonically decreasing, since a switch
turns an increasing substring into a decreasing one. Note that either n is involved in a simple switch, in which case there is nothing further to prove, or else
each switch involving n has length at least three.
If there have been no switches to the right of n, the length of the longest
monotonically increasing substring to the right of n is at most two, which can
happen only at the end of one substring and the beginning of another formed by
the switches involving n. Thus, the first switch involving elements to the right
of n in the permutation has a length of exactly two.
There must be at least one such switch, since:
(i) n must switch at least twice as there is no switch of length n, which corresponds to the case when all the points are collinear, an excluded case.
(ii) We have assumed that n switches before 1, implying that 1 is not involved
in the first switch involving n, which in turn implies that the elements to the
right n are not always monotonically decreasing.

References
[Baston and Bostock 1978] V. J. Baston and F. A. Bostock, A Gallai-type problem,
J. Combinatorial Theory Ser. A 24:1 (1978), 122125.
[Beck 1983] J. Beck, On the lattice property of the plane and some problems of Dirac,
Motzkin and Erd
os in combinatorial geometry, Combinatorica 3 (1983), 281297.
[Borwein 1982] P. Borwein, On monochrome lines and hyperplanes, J. Combin.
Theory Ser. A 33:1 (1982), 7681.
[Borwein and Edelstein 1983] P. Borwein and M. Edelstein, A conjecture related to
Sylvesters problem, Amer. Math. Monthly 90 (1983), 389390.
[de Bruijn and Erd
os 1948] N. G. de Bruijn and P. Erd
os, On a combinatorial
problem, Nederl. Akad. Wetensch., Proc. 51 (1948), 12771279 = Indagationes
Math. 10 (1948), 421423.
[Chakerian 1970] G. D. Chakerian, Sylvesters problem on collinear points and a
relative, Amer. Math. Monthly 77 (1970), 164167.

PROBLEMS RELATED TO THE SYLVESTERGALLAI THEOREM

493

[Clarkson et al. 1990] K. L. Clarkson, H. Edelsbrunner, L. J. Guibas, M. Sharir, and E.

Welzl, Combinatorial complexity bounds for arrangements of curves and spheres,
Discrete Comput. Geom. 5:2 (1990), 99160.
[Coxeter 1948] H. S. M. Coxeter, A problem of collinear points, Amer. Math. Monthly
55 (1948), 2628.
[Crowe and McKee 1968] D. W. Crowe and T. A. McKee, Sylvesters problem on
collinear points, Math. Mag. 41 (1968), 3034.
[Csima and Sawyer 1993] J. Csima and E. T. Sawyer, There exist 6n/13 ordinary
points, Discrete Comput. Geom. 9:2 (1993), 187202.
[Dirac 1951] G. A. Dirac, Collinearity properties of sets of points, Quart. J. Math.
Oxford (2) 2 (1951), 221227.
[Dirac 1959] G. A. Dirac, review of [Kelly and Moser 1958] in Mathematical Reviews
(MR20:3494), 1959.
[Edelsbrunner and Welzl 1985] H. Edelsbrunner and E. Welzl, On the number of line
separations of a finite set in the plane, J. Combin. Theory Ser. A 38:1 (1985),
1529.
[Erd
os 1943] P. Erd
os, Problem 4065, Amer. Math. Monthly 50 (1943), 169171.
[Finschi and Fukuda 2003] L. Finschi and K. Fukuda, Combinatorial generation of
small point configurations and hyperplane arrangements, pp. 425440 in Discrete
and computational geometry, edited by B. Aronov et al., Algorithms Combin. 25,
Springer, Berlin, 2003.
[Fukuda 1996] K. Fukuda, question raised at the problem session at the AMSIMS
SIAM Joint Summer Research Conference on Discrete and Computational Geometry
(South Hadley, MA), 1996.
[Goodman and Pollack 1980a] J. E. Goodman and R. Pollack, On the combinatorial
classification of nondegenerate configurations in the plane, J. Combin. Theory Ser.
A 29:2 (1980), 220235.
[Goodman and Pollack 1980b] J. E. Goodman and R. Pollack, Proof of Gr
unbaums
conjecture on the stretchability of certain arrangements of pseudolines, J. Combin.
Theory Ser. A 29:3 (1980), 385390.
[Goodman and Pollack 1984] J. E. Goodman and R. Pollack, Semispaces of configurations, cell complexes of arrangements, J. Combin. Theory Ser. A 37:3 (1984),
257293.
[Gr
unbaum 1970] B. Gr
unbaum, The importance of being straight, pp. 243254 in
Proc. Twelfth Biennial Sem. Canad. Math. Congr. on Time Series and Stochastic
Processes, Convexity and Combinatorics (Vancouver, B.C., 1969), Canad. Math.
Congr., Montreal, Que., 1970.
[Gr
unbaum 1972] B. Gr
unbaum, Arrangements and spreads, CBMS Regional Conf.
Series in Math. 10, Amer. Math. Soc., Providence, RI, 1972.
[Hanani 1951] H. Hanani, On the number of straight lines determined by n points,
Riveon Lematematika 5 (1951), 1011.
[Kelly and Moser 1958] L. M. Kelly and W. O. J. Moser, On the number of ordinary
lines determined by n points, Canad. J. Math. 1 (1958), 210219.

494

NIRANJAN NILAKANTAN

[Kelly and Rottenberg 1972] L. M. Kelly and R. Rottenberg, Simple points in

pseudoline arrangements, Pacific J. Math. 40 (1972), 617622.
[Lang 1955] G. D. W. Lang, The dual of a well-known theorem, The Mathematical
gazette 39 (1955), 124135.
[Levi 1926] F. Levi, Die Teilung der projektiven Ebene durch Gerade oder Pseudogerade, Berichte u
ber die Verhandlungen der K
oniglich S
achsischen Gesellschaft der
Wissenschaften zu Leipzig. Math-Physik Classe 78 (1926), 256267.

[Melchior 1941] E. Melchior, Uber

Vielseite der projektiven Ebene, Deutsche Math.
5 (1941), 461475.
[Motzkin 1951] T. Motzkin, The lines and planes connecting the points of a finite
set, Trans. Amer. Math. Soc. 70 (1951), 451464.
[Motzkin 1967] T. S. Motzkin, Nonmixed connecting lines, Notices of the American
Mathematical Society 14 (1967), 837837.
[Pach and Pinchasi 2000] J. Pach and R. Pinchasi, Bichromatic Lines with Few
Points, J. Combin. Theory Ser. A 53 (2000), 326335.
[Steinberg et al. 1944] R. Steinberg, R. C. Buck, T. Gr
unwald, and R. E. Steenrod,
Solution to Problem 4065, The American Mathematical Monthly 51 (1944), 169
171.
[Sylvester 1893] J. J. Sylvester, Mathematical Question 11851, Educational Times
59 (1893), 385394.
[Szemeredi and Trotter 1983] E. Szemeredi and W. T. Trotter, Jr., Extremal problems
in discrete geometry, Combinatorica 3:3-4 (1983), 381392.
[Tingley 1975] D. Tingley, Monochromatic lines in the plane, Math. Mag. 48:5
(1975), 271274.
[Ungar 1982] P. Ungar, 2N noncollinear points determine at least 2N directions, J.
Combin. Theory Ser. A 33:3 (1982), 343347.
[Welzl 1986] E. Welzl, More on k-sets of finite sets in the plane, Discrete Comput.
Geom. 1:1 (1986), 95100.
[Williams 1968] V. C. Williams, A proof of Sylvesters theorem on collinear points,
Amer. Math. Monthly 75 (1968), 980982.
Niranjan Nilakantan
Computer Science Department
Courant Institute
New York University
421 Warren Weaver Hall
251 Mercer Street
New York, NY 10012
United States
nilakant@cs.nyu.edu